Uma comparação entre abordagens de IA para o jogo Risk

Abstract

Over the last few decades, games have proven to be great test environments in the artificial intelligence field due to its well defined rules and clear evaluation methods. Therefore, this monography proposes the development, analysis and comparison of agents for the game Risk, a famous strategy board game based on turns, aiming the advance in the artificial intelligence field. For that, the following methods will be proposed to the agents development: heuristic and Monte Carlo tree search. The agents are going to face each other in two player matches, documenting the bots’ win rate against each of their opponents and the average number of troops owned by each agent at the end of the game. The expectation is to identify the pros, cons and efficiency of each trained model, concluding at the end which agent has the best performance.